[part of Simon Grant's research]
Modelling and simulating cognition,
and its relevance to complex tasks where humans are in control
Slides for the Grantholders days presentation
Modelling and simulating cognition, and its relevance to
complex tasks where humans are in control
Both I, for the last year and a quarter, and the work,
for the last few years, have been in ISEI, Industry Environment Unit,
Socio-Technical Systems Safety Sector (or their predecessors).
Several other grantholders have done probably
the majority of the work over those years.
Motivation
Accidents happen.
-
nuclear power plant
-
chemical industry
-
aviation
-
shipping
-
traffic control
Increasingly, the technology operates as it was designed to operate, and either
the human operation or design is mistaken.
Why?
[Gary Larson cartoons used as illustrations here]
Logical connections
The aim to reduce accidents.
One option is to reduce human mistakes.
This requires understanding of human mistakes.
Assume that mistakes arise from the same mechanisms as useful behaviour.
Then, the aim becomes to model and simulate human cognition and behaviour in a
way that will better enable
-
good human factors training
-
the analysis of human causal factor in accidents
-
the evaluation of human and joint system reliability
-
the design of human-centred automation, including interfaces and safety
features
-
the analysis of human information needs for the development of complex software
This is difficult!
What has been done?
The main centre of effort has been COSIMO, though there have also been related
projects.
COSIMO is a cognitive simulation model based on the psychological work of James
Reason.
It has involved several years of grantholder work, together with some stagiares,
and agents.
It has been the object of several visits and visiting scientists, including
eminent ones in the field of cognitive systems engineering.
COSIMO is interesting, but of course far from perfect.
What is being done?
The main topics with active effort are three:
-
modelling commercial aircraft piloting skill is related to Crew Resource
Management training courses;
-
co-operation and stress are subjects of interest in the domain of Air Traffic
Control;
-
development of architecture for modelling and simulation works towards more
efficient and effective modelling of cognition in any complex task.
What could be done?
KEY:
- Green blobs - processes
- Blue boxes - products
- Red arrows - forward development process
- Orange arrows - feedback
There are a number of useful activities that
can be traced out on this diagram.
- Observe; record; simulate and test loop. We can imagine
this being done with some cognitive simulations, e.g. Amalberti.
- Theorise, architecture, model build loop.
This is relatively easy.
- Theory-driven model and simulation building.
Maybe Soar and ACT-R are a little like this.
-
What else could be done?
Modelling and simulation of cognition could contribute to a wide range of
current activities:
-
human factors training courses;
-
accident analysis;
-
risk and reliability analysis;
-
the study and evaluation of expert judgement;
-
the design of robust human-machine interaction;
-
the development of advanced software tools;
-
human understanding of and interaction with complex equipment and natural
systems.
All of these may be expected to continue to be present in the new ISIS, and the
last ones probably in a large proportion of the JRC.
The future for modelling of cognition
Practical political future in the JRC.
ISIS seems an excellent place for:
-
some work in the field of nuclear safety;
-
increasing amounts of work in transport safety, perhaps including safety
systems, licensing, regulations and standards.
[should it be 'hazards' or should it be analysis of systems?]
The intellectual future in a wider context:
-
better simulation architectures for cognition
-
better modelling techniques usable in many ways